Diagnosis of Post Traumatic Stress Disorder through Speech-Based Emotion Modelling with 1D CNN

In contemporary society, Post Traumatic Stress Disorder (PTSD) continues to carry a stigma, as it denotes an individual who struggles to move beyond a concluded traumatic event. This study uses Deep Learning algorithm to tackle PTSD from the perspective of emotion recognition. An assessment instrument typically used by a psychiatrist to make the diagnosis of PTSD is a survey form. Scores derived from questionnaire-based assessments are deemed unreliable, and they have the potential to induce anxiety, distress, and discomfort among patients. Non-stationary signals, which include speech signals, are the most widely used biomarkers of emotion recognition. Mel spectrogram features are derived from emotional speech from the RAVDESS database for this study. The 1D CNN classifier’s performance metrics are calculated for each emotion. The findings indicate that the straightforward 1D CNN prediction model surpasses current alternatives, demonstrating superior performance of classification accuracy with average of 88.54% for the diagnosis of PTSD.